RESEARCH ARTICLE Contrasting spatial, temporal and environmental patterns in observation and specimen based species occurrence data James D. M. Speed*, Mika Bendiksby, Anders G. Finstad, Kristian Hassel, Anders L. Kolstad, Tommy Prestø Department of Natural History, NTNU University Museum, Norwegian University of Science and Technology, Trondheim, Norway * [email protected] a1111111111 a1111111111 Abstract a1111111111 a1111111111 Species occurrence data records the location and time of an encounter with a species, and a1111111111 is valuable for many aspects of ecological and evolutionary analyses. A key distinction within species occurrence data is between (1) collected and preserved specimens that can be tax- onomically validated (i.e., natural history collections), and (2) observations, which are more error prone but richer in terms of number and spread of observations. In this study we ana- OPEN ACCESS lyse the distribution in temporal, spatial, taxonomic and environmental coverage of speci- Citation: Speed JDM, Bendiksby M, Finstad AG, men- and observation based species occurrence data for land plants in Norway, a region Hassel K, Kolstad AL, Prestø T (2018) Contrasting with strong climatic and human population density gradients. Of 4.8 million species occur- spatial, temporal and environmental patterns in observation and specimen based species rence records, the majority (78%) were observations. However, there was a greater species occurrence data. PLoS ONE 13(4): e0196417. richness in the specimen record (N = 4691) than in the observation record (N = 3193) and https://doi.org/10.1371/journal.pone.0196417 most species were recorded more as specimens than observations. Specimen data was on Editor: Ulrich Joger, State Museum of Natural average older, and collected later during the year. Both record types were highly influenced History, GERMANY by a small number of prolific contributors. The species most highly represented in the obser- Received: December 13, 2017 vation data set were widespread or invasive, while in the specimen records, taxonomically Accepted: April 12, 2018 challenging species were overrepresented. Species occurrence records were unevenly spa- tially distributed. Both specimen and observation records were concentrated in regions of Published: April 26, 2018 Norway with high human population density and with high temperatures and precipitation, Copyright: © 2018 Speed et al. This is an open but in different regions within Norway. Observation and specimen records thus differ in taxo- access article distributed under the terms of the Creative Commons Attribution License, which nomic, temporal, spatial and environmental coverage for a well-sampled group and study permits unrestricted use, distribution, and region, potentially influencing the ecological inferences made from studies utilizing species reproduction in any medium, provided the original occurrence data. The distribution of observation data dominates the dataset, so inferences author and source are credited. of species diversity and distributions do not correspond to the evolutionary or physiological Data Availability Statement: All relevant data is knowledge of species, which is based on specimen data. We make recommendations for available from public repository (GBIF Occurrence users of biodiversity data, and collectors to better exploit the complementary strengths of DownloadÐ6th October 2017, doi:10.15468/dl. f2guqo. 2017). these distinct biodiversity data types. Funding: The authors received no specific funding for this work. Competing interests: The authors have declared that no competing interests exist. PLOS ONE | https://doi.org/10.1371/journal.pone.0196417 April 26, 2018 1 / 17 Specimen and observation based species occurrence data Introduction Recent years have seen a huge expansion in the quantity of georeferenced species occurrence data worldwide [1]. The sources of such data commonly are citizen science observations, and digitized natural history collection data. During the current global biodiversity crisis this data has proved to be a valuable source of information regarding species distributions and habitat associations, responses to climatic changes (e.g. [2, 3]), and conservation efforts (e.g. [4, 5]). However, species occurrence data is subject to several weaknesses. First, there are biases in geographical coverage of species records, taxonomic biases and temporal biases (e.g. [6, 7]). Secondly, the lack of information about the observation process often makes it often impossi- ble to draw inference on detection probability and false absences, hence reliable inferences regarding species distributions and their relation to environmental drivers cannot readily be made [8]. Thirdly, species occurrence data may also be subject to direct errors (e.g. [9, 10]), despite ongoing efforts to correct these [11]. Errors or inaccuracies may occur in the geo- referencing (i.e. the occurrence was not actually present at the location indicated), temporal reference (error or lack of specificity in the date given to the record), or taxonomic error (a misidentified taxon, or an outdated identification due to taxonomic revisions). Although such errors are not widespread [12], the ecological inferences drawn from these data may be sub- stantially affected [7, 13, 14]. Within a relatively short timeframe, international initiatives for ecological data availability (e.g. the Global Biodiversity Information Facility, GBIF) have cre- ated opportunities for the use of large quantities of species occurrence data. While this opens new doors for research, it also makes it paramount that we understand the limitations in the data quality and make our best efforts to correct for potential biases in the data. There has been considerable discussion and research into differences in biases and errors between citizen science data and professionally collected species occurrence data (e.g. [15, 16]). However, much of this discussion does not acknowledge there is a distinction between observation and specimen (or sample) based occurrence records, which is crucial for aggre- gated species occurrence databases. Both specimen and observation datasets may be comprised of a mixture of professional or citizen science collections and structured or unstructured sam- pling [17]. However, when the occurrence of a species is documented with a preserved speci- men in natural history collections, the taxonomic classification is reproducible and traceable. Here, the physical sample can be re-examined, thereby allowing the taxonomic classification of these occurrences to be validated and updated following taxonomic revisions. In contrast, observation-only data, collected by trained natural historians or by citizen scientists alike, lacks physical specimens to back-up the record [18]. While photographs can be used to docu- ment observations, for many taxa, a sample is required to reliably identify a species for exam- ple using microscopy [19]. There are therefore concerns regarding the certainty and therefore utility of observation based species occurrence data [20, 21]. The utility of specimen data for macroecological research is limited due to spatial and taxo- nomic gaps [21]. Meanwhile, the quantity of observation-based species occurrence data has rapidly increased over the past few years [1]. Observation data has the advantage of lower required effort, and does not require the mortality or disturbance to organisms caused by col- lection of specimens [22]. In many datasets it is therefore comprises a higher number of biodi- versity records. Specimen data has the advantage of taxonomic transparency. In addition, natural history collections preserve specimens that can subsequently be used for georeferen- cing studies of genetics or evolution (e.g. [23, 24]), phylogeography [25], physiology (e.g. [26]) or phenology [27]. Yet, biases and low sample sizes within natural history collections have con- strained the application of specimen data in some contexts [28, 29], often reflecting the geo- graphic and taxonomic preferences of a small number of contributors to such collections. PLOS ONE | https://doi.org/10.1371/journal.pone.0196417 April 26, 2018 2 / 17 Specimen and observation based species occurrence data Some of these limitations could be resolved through developing links with observation data that has the advantage of a potentially far greater sampling effort, and wider geographical reach. Specimen-based species occurrence records can provide an up-to-date taxonomic refer- ence that can be used to validate observation based species occurrence data, in addition to pre- served samples for future genetic or physiological analyses. A prerequisite to this is to first understand how such datasets differ in time, space and quality. In this study, we set out with the objective of testing the concordance between specimen- and observation-based species occurrence data. We compare temporal, taxonomic, spatial and envi- ronmental distributions between these two record types using a dataset of well-sampled taxa in a well-sampled region, namely land plants in Norway. We test the hypotheses that observations are in temporal terms (H1) more recent than specimen records [1]. In taxonomic terms, we test the hypotheses that observations (H2) record a lower diversity of plants in terms of species richness and evenness [30], but (H3) are more closely related to taxonomic prevalence (i.e. more common species are recorded more frequently) and (H4) sampled by a larger number of recorders than